key: cord-0024291-watl3k8c
authors: Duprez, Frédéric; de Terwangne, C.; Bellemans, V.; Poncin, W.; Reychler, G.; Sorgente, A.; Cuvelier, G.; Mashayekhi, S.; Wittebole, X.
title: High-flow nasal cannula therapy, factors affecting effective inspired oxygen fraction: an experimental adult bench model
date: 2021-12-08
journal: J Clin Monit Comput
DOI: 10.1007/s10877-021-00784-z
sha: 598e958c9ecda05b29b005f69def2b885f6661a9
doc_id: 24291
cord_uid: watl3k8c

Oxygenation through High Flow Delivery Systems (HFO) is described as capable of delivering accurate F(iO2). Meanwhile, peak inspiratory flow [Formula: see text] ) of patients with acute hypoxemic respiratory failure can reach up to 120 L/min, largely exceeding HFO flow. Currently, very few data on the reliability of HFO devices at these high [Formula: see text] are available. We sought to evaluate factors affecting oxygenation while using HFO systems at high [Formula: see text] in a bench study. Spontaneous breathing was generated with a mechanical test lung connected to a mechanical ventilator Servo-i®, set to volume control mode. Gas flow from a HFO device was delivered to the test lung. The influence on effective inspired oxygen fraction of three parameters (F(iO2) 0.6, 0.8, and 1, [Formula: see text] from 28 to 98.1 L/min, and HFO Gas Flows from 40 to 60 L/min) were analyzed and are reported. The present bench study demonstrates that during HFO treatment, measured F(iO2) in the lung does not equal set F(iO2) on the device. The substance of this variation (ΔF(iO2)) is tightly correlated to [Formula: see text] (Pearson’s coefficient of 0.94, p-value < 0.001). Additionally, set F(iO2) and Flow at HFO device appear to significatively affect ΔF(iO2) as well (p-values < 0.001, adjusted to [Formula: see text] ). The result of multivariate linear regression indicates predictors ([Formula: see text] , Flow and set F(iO2)) to explain 92% of the variance of delta F(iO2) through K-Fold Cross Validation. Moreover, adjunction of a dead space in the breathing circuit significantly decreased ΔF(iO2) (p < 0.01). The present bench study did expose a weakness of HFO devices in reliability of delivering accurate F(IO2) at high [Formula: see text] as well as, to a lesser extent, at [Formula: see text] below equivalent set HFO Flows. Moreover, set HFO flow and set F(IO2) did influence the variability of effective inspired oxygen fraction. The adjunction of a dead space in the experimental set-up significantly amended this variability and should thus be further studied in order to improve success rate of HFO therapy.

Oxygenation through high flow delivery (HFO) systems has been gaining increasing interest for the last 10 years among intensivists and emergency care physicians, especially in the context of advanced acute respiratory failure resistant to low flow oxygen therapy [1, 2] . The HFO devices are described as capable of delivering reliable, reproducible and accurate F iO2 from 0.21 to 1.0 with flows between 20 and 60 L/min [3, 4] . Hypoxemic critically ill patients do benefit the most of high flow oxygen devices and present with high Minute Ventilation ( V E ) and consequently high peak inspiratory flows ( V I ) [5] . This is especially the case of critically ill patients due to the Coronavirus 2019 disease (SARS-CoV-2). Indeed, despite severe pulmonary damage, the latter patients can maintain quite normal elastic-resistive thoraco-pulmonary characteristics during the first week of respiratory failure [6] thus being often capable of maintaining high V E and V I during a prolonged time [7] . In some cases, the peak inspiratory flow of patients with acute hypoxemic respiratory failure can even reach up to 120 L/min [8] , largely exceeding HFO flow. At these high V I , the assumption is made that atmospheric room air might enter the breathing circuit and dilute the set F iO2 from HFO [3] reducing consequently the effectively delivered F IO2 .

We firmly believe that a thorough understanding of F iO2 stability during HFO therapy is paramount as it is increasingly used in acute respiratory failure severity scores [9] [10] as well as scores predicting the need for intubation [11] . Since HFO is used in very sick hypoxemic patients, with high Respiratory frequency (Rf) and high Tidal Volumes (Vt) due to high oxygen demands and thus approximate V I up to 100 L/min [12] , such as critically ill SARS-CoV-2 patients, we sought to evaluate the reliability of HFO systems at these higher V I . Therefore, the aim of the study was to analyze the effect of high V E and V I on effective F IO2 during HFO treatment in an experimental adult bench model.

Additionally, since previous studies have found that the addition of a surgical mask or a Double Trunk Mask (DTM) (dead space) above HFO nasal cannulas may increase the arterial pressure in oxygen in some hypoxemic patients despite any modification in HFO settings [13, 14] , the secondary aim of the study was to evaluate the impact of the addition of a dead space on effective inspired F iO2 .

The HFO was generated by an Airvo2 ™ device (Fisher & Paykel Health Care, Auckland, New Zealand) through a T piece connector (INTERSURGICAL T-Piece Connector iso 22-DS00586). The T piece was connected to a mechanical test lung through a single ventilator tubing (iso 22). The first lung was driven by a mechanical ventilator (Servoi™ Maquet, Getinge group, Getingue, Sweden) (Labelled as Drive Mechanical Ventilator) to simulate a spontaneous breathing pattern.

Gas flow from Airvo2 ™ was set at 40, 50 and 60 L/min and confirmed by a calibrated flowmeter (SP-304 flow sensor-Iworx®, United States). F iO2 was set at 0.6, 0.8 and 1 and confirmed by a calibrated oxygen analyzer (two point calibration, GA-200 Side stream, Gas Analyzer Iworx®, United States) which was connected to a data-acquisition system IX-214 hardware Oxygen analyzer GA-200, first calibrated with room air (21%) and secondly at 100% with certified O 2 gas coming from an oxygen gas cylinder. To evaluate the dynamic variability of effective inspired oxygen fraction into this bench model during respiratory cycle, the measures of F iO2 were performed at two different locations:

(1) Directly into the artificial lung through the O 2 port of the second lung ( Fig. 1 ) (2) Near the T piece ( Fig. 1 ).

When an experimental setting was changed, it was left stabilizing for 3 min prior to measuring. The F IO2 was measured for 1 min and data of the final 3 breaths were extracted. The mean of these three measures is reported.

Spontaneous breathing was generated in Ambient Temperature and Pressure Saturated (ATPS) conditions with a mechanical test lung (Dual Test Lung-Michigan Instruments, Inc. Grand Rapids Model 5600i) including two independent artificial lungs. The two lung compartments were synchronized with a lung coupling clip, which allowed the first lung to drive the second lung in order to achieve spontaneous breathing simulation. Resistive and Elastic characteristics of Dual Test Lung were: 5 cmH 2 O/L/sec and 0.06 L/cmH 2 O.

The first lung was driven by a mechanical ventilator Servo-i™ (Servo i™ Maquet, Getinge group, Getinge, Sweden) set to volume control mode (descending ramp flow waveform without auto-flow, time pause and inspiratory rise time at 0%, peep of 0 cmH 2 O, the trigger was set at -10 cmH 2 0 in order to avoid self-triggering).

Respiratory frequency ( 

V E was defined as the product of Vt and Rf. Difference in measured F iO2 (mF iO2 ) and set F iO2 (sF iO2 ) was defined as ΔF iO2 . Continuous variables are presented as mean (± SD). The assessment of the relationship between V I and ΔF iO2 was computed through Pearson's correlation analysis in function of the different variable groups (set HFO Flow and sF iO2 ). To define which variables were the most determinant in predicting effective inspired oxygen concentration and ΔFi O2 , forward stepwise multiple linear regression was computed. Variables with a p-value < 0.01 were entered into the model. Multiple linear regression modelling was computed to evaluate the effects of selected variables ( V E , HFO Flow, and sFi O2 ) on ΔF iO2 and effective inspired oxygen fraction. The model was then internally validated with a 10-Fold Cross Validation Procedure. Differences between groups (with and without Trunk, sFi O2 and HFO Flow) were compared through ANCOVA adjusted for V I . Normality was assessed using the Shapiro-Wilk test and linearity between the dependent variable (ΔF iO2 ) and the covariate ( V I ) was assessed by correlation plots with Pearson's correlation coefficient per studied factor (Trunk, set HFO Flow and sF iO2 ).

Throughout analysis, a p-value below 0.01 was considered as statistically significant.

Statistical analysis was performed using R: A Language and Environment for Statistical Computing, R Core Team, Vienna, Austria, 2020, https:// www.r-proje ct. org/.

(Used packages emmeans, ggpubr, rstatix, lattice, car, caret).

The influence of the different variables on ΔF iO2 is demonstrated in Fig. 2 ; Tables 1 and 2. The experimental set up demonstrated high variation in measured F iO2 (mF iO2 ) through the respiratory cycle in function of the location of the measurement. At the T-piece, measured F iO2 oscillated from ± 10% whereas, at the lung sampling point, this oscillation was around 2-4%. Therefore, further FiO 2 measurements were conducted intra-pulmonary. The difference (ΔF iO2 ) between measured F iO2 and sF iO2, is influenced by V I , set Flow and F iO2 at HFO ( Fig. 4 ; Table 2 ).

V I exerted a positive linear relation with ΔF iO2 throughout the different variable groups (set HFO Flows and sF iO2 ) with a minimal Pearson's coefficient of 0.94 (p-value < 0.001). The higher V I , the more ΔF iO2 increases. This trend is accentuated with the increase of sFi O2 and the decrease of HFO Flow (Fig. 3) , reaching a maximum ΔF iO2 of 44% at V I of ± 98 L/min, sFi O2 at 1 with a HFO flow of ± 40 L/min. For each curve without trunk reported in Fig. 3 , there seems to be a threshold under which ΔF iO2 variations are less important.

This threshold matches the point at which V I equalize HFO Flow (intersection of dashed line and without trunk curve in Fig. 3 ).

Stepwise selection model (Table 1) showed V I to be the most influencing variable of ΔF iO2 . The model did improve significantly with the adjunction of HFO Flow and set F iO2 : R 2 increases, and a Root-Mean-Square Error (RMSE) decreases at each step.

The model, with ΔF iO2 as dependent variable, based on the stepwise selection model (Table 2 ) yielded a R 2 of 0.849 and a RMSE 3.83. This could be improved with a 10-Fold Cross Validation procedure to a R 2 of 0.92 and a RMSE of 3.76. All three variables ( V E , sF iO2 , and HFO Flow) are highly significant in predicting ΔF iO2 (p < 0.001).

Effects on ΔF iO2 of HFO Flow and sF iO2 , adjusted to V I (estimated marginal means), is shown in Fig. 4 . Set F iO2 significantly modified ΔF iO2 (ANCOVA adjusted for V I , F = 35.45 p < 0.001). On the other hand, HFO Flow also did significantly change ΔF iO2 (ANCOVA adjusted for V E , F = 48.14, p < 0.001).

When a trunk was added to the T-piece, ΔF iO2 did significantly decrease at each level of sF iO2 (Tukey's adjusted p-value < 0.01) (Fig. 5A) . ΔF iO2 did also decrease at each different set Flow at high oxygenation device with adjunction of a trunk (Tukey's adjusted p-value < 0.01) (Fig. 5B ).

The present bench study demonstrates that during HFO treatment, measured F iO2 in the lung does not always equal set F iO2 on the device. The substance of this variation At first, this study shows important oscillations in F iO2 measures through the respiratory cycle in function of the location of the measurements in the experimental set-up ( Fig. 1) . To simulate physiological conditions, i.e. open mouth or inevitable leaks around the nasal cannula, a T-piece was mounted in the experimental set up. F iO2 measurements at the T-piece present oscillation of up to 10% in F iO2 through the respiratory cycle in comparison with oscillations of up to 4% of "intra-pulmonary" measurements ( Fig. 1) . This first finding demonstrates presence of air blending in this part of the circuit, that stabilizes when arriving in the lungs. Analysis in the present study was therefore carried out on stabilized intra-pulmonary F iO2 measurements, as recommended by the manufacturer of the Dual Test Lung. Through the analysis of these measurements, important differences were found between sFiO 2 and mF iO2 (sF iO2 -mF iO2 = ΔF iO2 ) at different V I , Flow and set F iO2 . V I was the most significant variable influencing ΔF iO2 and exerted a positive linear relation with the latter. This finding accentuates the purpose of our study which assessed the effects of high V E and V I while other studies did limit this variable to physiologic "resting" values [3] .

Meanwhile, there seems to be a threshold of V I under which ΔF iO2 does diverge less. It was indeed previously hypothesized that once V I do exceed HFO Flow of the High Flow Oxygenation Device, air blending between high oxygenated air from HFO and room air from the mouth or leaks do occur and could thus decreases actual delivered F iO2 [3] . When V I does exceed HFO Flow, depression happen inside the circuit, which consequently causes a suction effect inside the upper airways. Entrained room-air arriving from the mouth or leaks around the nasal canula (the T piece in our experimental set up), blend with oxygenated air from HFO, thus decreasing F iO2 of inspired HFO flow.

From our data we can indeed extrapolate a breaking point corresponding to the point where V I and HFO Flow equalizes. This point can be found at the intersection of the dashed line ( V I = HFO Flow) and the curve "Without Trunk" in Fig. 3 . If above this point ΔF iO2 looks clear and was expected, it seems however that below this point, air blending does also occur.

To explain the latter, we make the assumption of a possible Bernoulli effect in the upper airways during inspiration when V I does not exceed HFO Flow. Laminar flux creates depression in the circuit and thus an air suction effect coming from the T-piece (anatomically comparable with the mouth or leaks around nasal cannula). Air blending will occur and generate a ΔF iO2 which value will vary on sF iO2 and HFO Flow (Fig. 3.) . Indeed, ΔF iO2 will increase when sF iO2 is set at higher values, and at lower HFO Flows.

The rational principle behind these different findings lays on the mathematical air-mixing equation of an air circuit [15] . With the formula of this principle, we simulated and depicted in Fig. 6 the relationship of F iO2 calculated through different sF iO2 at different Air Flows. Air Flow in this simulation illustrates the flow arising from the suction effect during either assumed Bernoulli effect either V I exceeding HFO Flow. As the mean difference in V between trunk and without trunk is statistically significant (p < 0.01) with Tukey's adjusted p-values and set HFO Flow was of 13L/min in our experiment, we choose to represent Air Flows of 5, 12.5 and 20 L/min. In this simulation, we see that the lower sF iO2 , the less it will be affected by room air blending, and thus, the difference in sF iO2 and calculated F iO2 will remain small. In the same way, the higher the Air Flow, the more air blending occurs, and thus, increasing difference between sF iO2 and calculated F iO2 . This is in accordance with our findings that higher sF iO2 and lower HFO flows tend to augment ΔF iO2 . Clinical applications of our findings are important. Indeed, how sicker the patients, the more important V I and the need of high F iO2 , the less reliable F iO2 delivery from HFO become. The only parameter limiting differences in sF iO2 and actual F iO2 in patients with high V I seem to be the flow rate at HFO device. We could therefore suggest augmenting flow in these patients before trying to augment F iO2 to diminish room-air suction effect of patients with major peak inspiratory flow. The use of devices with the ability to deliver higher Flows could also be determinant in assuring reliable F iO2 to patients with high V I . Another objective of our study was to analyze the effect of the adjunction of a dead space in the circuit. The addition of a tube ISO 22 of 30 cm of length (0.230 L) on the T-piece did significantly decrease ΔF iO2 . This could be explained by the fact that during expiration, high oxygenated air from HFO will be contained in the dead spaces (mouth, surgical mask or Double Trunk Mask). According to the air-mixing theorem, during inspiration, with Inspiratory Flow above HFO Flow, instead of room air at a fraction of 0.21 of oxygen, air at higher oxygen fractions will blend with high oxygenated air from HFO, thus limiting ΔF iO2 . Hence, adjunction of a dead space (Double Trunk Mask or equivalent) could also be a promising premise to enhance HFO therapy as it was already clinically demonstrated [13, 14] and should be further studied.

The present study has limitations. In clinical practice it is difficult to appreciate actual V E and estimate the T i on total respiratory cycle time (T i + T e ) ratio which determine mean inspiratory flow and could help estimating peak inspiratory flow. Furthermore, as it is a bench study, the clinical consequences can only be hypothesized and warrant further in vivo research. Our model does not exactly reproduce three-dimensional anatomy of the upper airways. Nevertheless, in physiological situation, we expect less ideal gas flows than our experimental design. Hence, we expect more turbulent flows in patient's upper airways, which only can augment our estimate ΔF iO2 through amplifying air blending.

The present bench study did expose a weakness of HFO devices in reliability of delivering accurate effective inspired F IO2 at high V I as well as, to a lesser extent, at V I below equivalent HFO Flows. Moreover, set HFO flow and sF IO2 did influence variability of effective inspired oxygen fraction. Until further in vivo research of stability and reliability of effective inspired oxygen fraction during HFO treatment, we advocate to use sF IO2 with caution in severity scores and scores predicting the need for intubation.

Nevertheless, the adjunction of a dead space in the experimental set-up significantly amended variability between sF IO2 and effective inspired F iO2 and could thus be a promising start point to improve success rate of HFO therapy.

Author's contribution CT, FD, WP designed the study, acquired data and wrote the manuscript. Profs XW, GR, AS and GC wrote, read and approved the final manuscript. Drs CT, SM and VB have contributed to bench study conception and statistical calculation.

The study did not receive any funds of any kind. 

High-flow oxygen administration by nasal cannula for adult and perinatal patients

High-flow nasal cannula oxygen therapy in adults

FIO2 in an adult model simulating high-flow nasal cannula therapy

High-flow nasal cannula oxygen therapy in adults: physiological benefits, indication, clinical benefits, and adverse effects. Respir Care

Inspiratory work with and without continuous positive airway pressure in patients with acute respiratory failure

COVID-19 does not lead to a "Typical" acute respiratory distress syndrome

Pre-hospital critical care management of severe hypoxemia in victims of Covid-19: a case series

Physiologic effects of noninvasive ventilation during acute lung injury

The SOFA (Sepsis-related Organ Failure Assessment) score to describe organ dysfunction/failure. On behalf of the Working Group on Sepsis-Related Problems of the European Society of Intensive Care Medicine

Acute respiratory distress syndrome: the Berlin Definition

An index combining respiratory rate and oxygenation to predict outcome of nasal high-flow therapy

Analysis of hypoxic and hypercapnic ventilatory response in healthy volunteers

The double-trunk mask improves oxygenation during high-flow nasal cannula therapy for acute hypoxemic respiratory failure

Surgical mask on top of high-flow nasal cannula improves oxygenation in critically ill COVID-19 patients with hypoxemic respiratory failure. Ann Intensive Care

Ambient pressure air/oxygen blender

Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations